Oh oh it's OCZ again 270FSB

[Liquid3D]

It's getting better all the time. :) Got my Gold 4000 today. IGB-kit (2x512MB) I threw the memory in, finally flashed the Epox 4PCA3+ BIOS, at 275FSB, 1:1, SPD, and 2.9V, she's been running more stable then a codependent lover with an insecurity complex.

What's odd is there's only one BIOS update for 4PCA3+. At the USA site it's description is one line; "Fixes sound crackling...". At the Euro Epox site the description is three lines, of which one states; "enables more aggresive memory settings." Well there's exactly two PAT options with the new BIOS, and one is CPC which needs to be disbaled in 90% of the motherboords. What's perplexing to me, is how can Epox ship th board without any PAT settings if it's an 875 mobo? Anyway the benchmarks below were run with PAT features diabled, Don't want to overwork my girls. After all a good product "pimp" keeps his stable healthy. :D

Stay tuned for more trickin out the RAM action.
EDIT: PAT features are disabled in this test

[Major]

Awsome !!!!:toast: gota love 270mhz fsb 1:1 !!!

One question, you have so many toys listed in your sig ! hehe which OCZ where you using ??

Edit: and what timings and Vdimm ??

2nd Edit: ok I see 3-4-4-8 hehe

[eva2000]

i'm about to install my 4PCA3+ rev 1.1 as well, do you mean only latest bios has PAT enabled or is it still disabled ? what does ctiaw and cpuz 1.19 report for memory/PAT ?

[STEvil]

Makes me wish I had bought different ram than this BH-6 XMS.... arghnuts! 8-)

[Shane5578]

fast cs should be enabled and cpc disabled.

otherwise cpuz/ctiaw will show PAT disabled

cpc only works near stock speeds for me

[CCW]

Nice bandwidth! Doing great, P4s are only proper way to test mem bandwidth. :toast:

Also, you sig is great :banana:


L310A735 Malay 2.4C@3.696MHz
Q320A287 Malay 2.4C@3.6GHz
JIUHB1700DUT3C@2.5GHz
Thermalright SLK900U
Vantec Tornado 92mm
Prometeia r404
Abit IS7-E
Abit IC7-G
Epox 4PCA3+
Epox 8RDA
Soltek SL75FRN2-L
Tyan Tachyon G9700 Pro 128MB
Maxtor SATA150 120GB
2x512MB OCZ EL-PC3700
4x256MB OCZ GolD DC3700
2x512MB Corsair Twin-X3200LL
2x512MB Corsair XMS3500
2x512MB Corsair Twin-X4000
4x256MB Corsair Twin-X4000
Antec True Blue 480W
PCPower&Cooling Silencer 400W

[SpicyHuevos]

i found timings of 3-3-4-6 works the best up to 290 1:1
May want to give it a shot:)

[althes]

DAmn I have to get some of those.:D

[Kalway]

CAS 3 again? No wonder your efficiency is crap. Or is it that PAT improves bandwidth that much more? 6.1gb/sec at 270mhz is really quite sad considering you can get more than that at 200mhz with lower latency. Sorry but high FSB really means nothing to me unless you have the actual bandwidth to back it up. That bandwidth looks like a couple of pc3200 sticks at 2.5-3-3-7. I wish I could show the bandwidth on our test rigs in the labs with 875 chipsets. Luckily the company I work at has the same philosophy on RAM and at least refuses to use above CAS 2.5 for stability and efficiency purposes.

[jaawood]

Originally posted by Kalway
6.1gb/sec at 270mhz is really quite sad considering you can get more than that at 200mhz with lower latency. Sorry but high FSB really means nothing to me unless you have the actual bandwidth to back it up. That bandwidth looks like a couple of pc3200 sticks at 2.5-3-3-7.

i agree that tighter timings would be better, but i'd be REALLY impressed to see that bandwidth at 200 mhz, i think that is impossible. also, his sandra bench clearly compares his score to PC 3200 cas 2.5, and more than doubles it. im not sure what you are thinking....:stick:

[Kalway]

That's Dual Channel is it not? 3200 x 2 = 6400 does it not? Sandra is comparing it to Single Channel. I beat those compares with my nForce2 at 195mhz. With 200mhz I get at least 95% efficiency and sometimes up to 97%. But like I said I only use timings of 2-2-2-5 because the mhz usually doesn't make up for the relaxed timings. You might gain about 30-50mb/sec by relaxing timings but usually that's pushing it to the memory's absolute max.

[KennethChong]

Very nice, im guna have to look into this memory.

[Major]

I have two sticks of OCZ Gold 3700 coming on next Tuesday via Fedex ! I hope I can do 270mhz ! thats so sweet !

[Kalway]

I still don't see what's so spectacular about that. You can get higher bandwidth at stock.

[Major]

If you have the same OC on processor, say 4ghz, with the 5:4 divider and your memory at 200mhz with tight timings like 2-5-2-2 your going to have some decent memory bandwith, but I feel that with 1:1 and your memory at 270mhz for the same OC of 4ghz even with loose timings like 3-4-4-8 overall system performance has to be better.

[KennethChong]

well i just orderd some, coudnt help myself :/ should be here by next week. I went with 2x 512, hopefully this memory should be able to get me to 4.3 with a gig, unlike my buffalo.

[mike.elmes]

Better overall system speed with loose timings will outperform lower speed and better bandwidth. Can someone bench a few 3dmark runs to prove??

[KennethChong]

Yes thats true. What i think major was saying is that if he can only get to 4 ghz with his current ram at 5:4, then if hes able to get the same oc with 1:1 that will be faster. Im not sure about the timings yet, but they should be decent with this mem. I also think that lower timings with better mhz is the way to go. At least with that much of a jump.

[Kalway]

Why bother going 1:1 when the divider is there to lower the memory clock to allow for tighter timings and better bandwidth? You're only crippling bandwidth using the 1:1 with loose timings. Use the 5:4 divider, (FSB:memory to make sure I'm clear), tighten the timings and you'll get the same cpu speed but most likely more bandwidth because the memory isn't timed like a grandfather clock. Using the divider your memory won't hold you back, you'll get more out of your cpu, and probably get better bandwidth.

I'm sure 250mhz 2-2-2-5, or even 2-3-3-7, on DCDDR will get better bandwidth than 270mhz 3-4-4-8. That's only 20mhz difference and a big difference in timings. You see what I'm saying?

[KennethChong]

From what ive heard and seen, 1:1 will get you better sandra scores, now when it comes to 3dpreformance it may hamper it, i havnt done the test myself, so im not sure about that. But yes it would help sandra. I dont only bench 3dmark :) And my reason for getting this mem is for the 5:4 divider, it will most likly offer me more than this buffalo is getting. But yes, at 1:1 you may indeed be right with 3d programs.

[STEvil]

It depends on each setup.

[Shane5578]

its preferable running 1:1 as there no memory bottleneck

remember your memory is bottlenecked as soon as you use lower dividers

[Liquid3D]

Originally posted by SpicyHuevos
i found timings of 3-3-4-6 works the best up to 290 1:1
May want to give it a shot:) Please show me 1:1 at 290FSB. I'm not doubting your honesty, I just really want to see how you did it, WoW.

PAT features were disabled for this initial test Below are my first results at 270FSB instead of 275.6FSB, PAT features were disabled as well.

In so far as Bandwidth is concerned. It's not difficult to figure where this memory would perform had the timings been slightly tighter. Since i haven't even tried to tighten them yet, we don't know what the memory will do. That's next. Right now i have some articles to finish. Although I do agree tighter latencies are a good thing, the bandwidth is still above 6000MB/s. Those running A-DATA or Twinmoss at tighter latencies, and such high FSB speeds need to push more then 2.99V into their memory. Given the fact there may not exist IC's which began life as a 3.5ns, or even 4ns speeds (as so many claim) these latencies are realistic. 3DMark performance is improved anytime your memory is stable, and I don't beleive any memory running at these speeds with tigher latencies is going to be stable for long, especially at 3.1V, or 3.3VDIMM. Memory designed for Canterwood's maximum performance doesn't need to be LL. Those memory aggressive features are innate, in the MCH itself. Which is why loose timings are required for stability, and especially in 3D applications. What's the difference if I'm still getting 6100MB/s bandwidth. Yes they "COULD" be tighter, and Intel "COULD" unlock it's multipliers. Any professional in the memory industry will tell you that wanting LL at DDR500 speeds is not being realistic. Not with Canterwood, unless your continually pumping unsafe voltage into the chips. How long do you think the memory, your system, will remain error free under such duress? Their designing for speed yes BUT MCH stability is the foundation upon which that speed runs. Their not designing thousahnds of modules for the 3% of us crazies who want 8000/MBs in a 6000MB/s world. You can't have it all. The Canterwood, and nForce-2/3 chipsets are different animals. One must also rmember this is Intel's first real, venture into a Dual DDR platform (if you doubt this read the White Papers on both Granite Bay and Canterwood). Given the fact most people have to disable PAT to attain very-high FSB speeds, indicates revisions are needed. My Pseudo-PATronizing... article predicted correctly the un-certain future of Hyperpath, GAT, and the other Pseudo-PAT features for 865. Intel announced two weeks ago revision-2 of the 865, will not allow any psudo-PAT type BIOS features n Springdale boards.

I'm the idiot savant. Would someone please tell me what CTIAW is, and where the Hey does CPU-Z indicate PAT?

[Kalway]

By the way theoretical bandwidth on Liquid3d's setup is ~8700mb/sec. He's getting ~6150mb/sec (rounding up for easier math). That's 70.6% efficiency. At 2-2-2-5 you get ~95% efficiency. My point is there is actually less of a bottleneck when using the divider since the memory is sending data faster. Run the memory at 200mhz with 2-2-2-5 DCDDR and see what your bandwidth is. It should be ~6000-6200mb/sec (Intel's chipsets are annoying with their efficiency).

IF my calculations are correct at 240 2-2-2-5 you SHOULD get easily over 7000mb/sec. If one of you with the memory + board would be kind enough to try this please post your results. Try 1:1 and 5:4 if you can also try with PAT enabled. I'd get some myself but there are no test boards available right now :( so I can't do it. We also don't have any memory that can reach that speed here haha.

Post with some results.

[NyCUndaGrounD]

How much are u guys paying for the OCZ PC4000. Im sad my Adata can do 267mhz 1:1, 24 hours afteri got them. Will try to burn in and see how far it can go.

[Shane5578]

clevor already tested 2-5-2-2 at 250 fsb on the 4pca3+

he got around 6100 but his unbuffered score was around 4000

so thats the same as 280fsb using slack timings. lol


oh i hear a-data pc4000 modules are good, they just need 3vdimm

[Kalway]

Wow 6100 at 250? That is horrible. Why is the Canterwood's efficiency so crappy? Opteron would blow it out of the water at those speeds, since it has the built in mem controller. Sort of reaffirms my proof that Intel is not building as quality chipsets as they should be to make all these bandwidth claims. So at 250 that should be 8gb/sec right? Even on tight timings that's only 76% efficiency. I think Intel needs to invest in some Hypertransport technology if we seem to be topping out at 6100.

[Shane5578]

lol

originaly DDR memory wasn't even meant to do 200fsb, so its quite amazing they got this far.

most likely DDR2 will take over

[Major]

You guys are missing the point !!

You can't tell me that running a system at 4ghz with fsb of 266 with 5:4 div running memory at 213mhz with tight timings is going to outperform the same system at 4ghz with fsb of 266 with 1:1 div running memory at 266mhz with regardless of timings. Thats my point !

Advantages: 1) no latency from running divider 2) 53mhz increase in ram speed [it would take more than going from cas 2 to 3 to make that up] 3) no latency from running divider 4) no latency from running divider ! hehe

[STEvil]

Originally posted by Kalway
By the way theoretical bandwidth on Liquid3d's setup is ~8700mb/sec. He's getting ~6150mb/sec (rounding up for easier math). That's 70.6% efficiency. At 2-2-2-5 you get ~95% efficiency. My point is there is actually less of a bottleneck when using the divider since the memory is sending data faster. Run the memory at 200mhz with 2-2-2-5 DCDDR and see what your bandwidth is. It should be ~6000-6200mb/sec (Intel's chipsets are annoying with their efficiency).

IF my calculations are correct at 240 2-2-2-5 you SHOULD get easily over 7000mb/sec. If one of you with the memory + board would be kind enough to try this please post your results. Try 1:1 and 5:4 if you can also try with PAT enabled. I'd get some myself but there are no test boards available right now :( so I can't do it. We also don't have any memory that can reach that speed here haha.

Post with some results.

That only boosts the total used of available bandwidth, though. Overall we want more data through and in a more timely matter, reguardless of FSB/timgings.


It all depends on each setup.

[Kalway]

I see... Yeah, there needs to be some more dividers actually to assist the memory along with the FSB. 5:4 is a little too much now that I think about it. I'm thinking larger dividers would probably be a lot more useful. 10:9 would probably be more useful than 5:4... I did my math on all of it except for the fsb:memory ratio. I geuss the main problem is really just the chipset's and the memory's maximum efficiency.

Looks like we are at the point where DDR2 needs to come in. DCDDR2 would be preferable and probably required to get beyond DCDDR1 speeds. Any word on when QDR is scheduled to be available?

[Clevor]

As Shane mentioned, I've done extensive testing of ADATA PC4000 at 3-4-4-5 (and even tighter timings) versus Corsair XMS 2500 C2 at 2-2-2-5. The Corsair wins out every time at 5:4 and tighter timings.

The 4PCA3+ does not get impressive buffered scores, but the unbuffered is exceptional. In Win98SE, I get 3900/3700 unbuffered at 250, 1:1, 2-2-2-5. Here are my unbuffered at same settings in WinXP, much lower, but it ain't easy getting above 3400 in WinXP:

[Clevor]

In every instance where I've compared Sandra unbuffered/3DMark between different rigs, the rig with the higher unbuffered got better 3DMark scores. I've found the PCMark2002 memory score is more influenced by buffered memory scores.

I've gotten my 4PCA3+ up to 282, 1:1, 3-4-4-5 on the 4PCA3+, by using 2.9 VDIMM. There's no doubt the ADATA PC4000 can hit DDR600 if you got a CPU that can go that high and if you use 3.1+ volts.

As Shane mentioned, if you enable 'Intel FAST' in the BIOS, that will leave on PAT. I have not found much performance difference with it on, maybe a 0.4% increase in 3DMark.

This is my unbuffered at 277, 5:4, 2-2-2-5 on the 4PCA3+, with the XMS3500. Pretty decent for DDR443:

[Liquid3D]

Your sad your A-DATA wil ONLY do 267FSB. That's pretty good. Can you post a quick Sandra screenshot.

Kalway i don't think this will run at tighter timmings even at slower speeds because it's SPD for 3-8-4-4, but I'll try. Did someone say they got better results lower FSB, and tighter timings? How do you think Corsair Twin-X3200LL would do? It didn't do that well on my IS7-E? But then again I think I was shooting for maximum CPU speed, not bandwidth. Should I try the TwinX3200LL? How low of a FSB?

[#ball3r]

liquid3d....check you pm's! ;). good results none the less

[Kalway]

Much better I'm seeing there. I need to reinstall Sandra on my rig at home. I need to see what my unbuffered bandwidth is on my NF7.

Overall, Clevor, you just reassured me of some evident problems I've seen with the Canterwood chipsets and that timings are extremely important in bandwidth, even on 875/865. Also, this proves that PAT is just a stupid marketting gimic that does practically nothing. Sort of helps our engineers really because now we don't have to retest some boards hehe.

Still curious what the buffered and unbuffered stock bandwidth is with and without PAT. If it's what I'm expecting it to be then we'll probably have a better idea what's causing the horrible efficiency on Canterwood's architecture. Wish I could get a true DCDDR bandwidth test on nForce2.

[saaya]

niiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiice liquid :) :toast:

[Clevor]

Overall, Clevor, you just reassured me of some evident problems I've seen with the Canterwood chipsets and that timings are extremely important in bandwidth, even on 875/865.

I compared the same fsb and DDR setting on the 4PCA3+. At 2-2-2-5 I got 3300+ unbuffered, at 2.5-3-3-6, I got 2750. 3DMark2001 dropped about 400 points in the 18k range.

I did check Sandra with PAT on and off once and I think I saw maybe 50 points increase with it on (4PCA3+).

Only other CW board I've tested is a P4C800-E. I could overclock 10 mhz higher with the same CPU (287), but it could run 0.05+ more VCORE. That board gave higher buffered scores compared to the 4PCA3+, but unbuffered was about 100-200 points lower.

[Zroc]

Purty scores, Liquid ;)
By the way, you can't disable PAT on an 875P chipset (what the BIOS is refering to is not actual PAT, but some other functions, similar to Game Acceleration on Abit, etc). PAT is on with an 875P regardless...the only way to shut it off is sticking and 865PE in there ;)

Kalway: you've got it backwards. When memory bandwidth is equal to FSB bandwidth on P4's, Sandra's 'efficiency' (that is such an off term) is always in the 70's (even back in RDRAM 4x and GB days) This is NOT telling you your setup is 'inefficient', like most people assume, but rather it's telling you that you actually don't have a bottleneck. Think of it as your FSB bandwidth having room to breath.

Now, when you use a divider to make your memory bandwidth LESS than your FSB bandwidth (or in the old days, single-channel), your 'efficiency' jumps into the 90's.
What this means is, you DO have a bottleneck...your FSB is using almost all your memory bandwidth.

Hard to think of it in those terms (a really high 'efficiency' really means your bandwidth setup is bottlenecked), but that's how it works.

[Liquid3D]

Thank you Kalway I felt as if I was the only crazy person or something trying to say there seems to some architectual problems with the 875's NB-MCH (Memory Controller Hub)! I first saw the signs of trouble when Pseudo-PAT features were released onto 865 boards. I've yet to see anyone truly exploit those features, and in fact this whole "speed-bining" process may be at the core of the issue. Intel set out to design the Canterwood, and the Springdale comes along because it's like the bastard child no one really wanted, but they were dammed if they were going to toss that many chips. All of sudden you realize there's no 3% to 5% differnce in performance, in fact what you see is many 865 boards outscoring certain 875 boards, and inconsistent performance gaps everywhere. Seems Asus and MSI whom, usually fall behind Epox and Abit in overclocking prowess, have had the best luck with the chipsets.

I made the claim, and adhere to it, that the 865, is the first indicator there's problems with the 875 NB-MCH. It's as if uit's conflicting with any and all memory thayt comes it way except, for a few exception. Whikle OCZ and Corsair, held the performnace crown with the nForce-2, it's now as if their struggling to find SPD, or IC's compatable with the MCH. Then comes along A-Data and Twinmoss who make good products, but where the Intel 875 is concerned, why are they the only brands to hold these tight timings, while the so-called gienats are producing memory which must be run at 3-7-3-3 or 3-8-4-4?

I guess the question is, what is A-DATA using, or doing which allows their memory to either circumvent the problems innate to the Canterwood's MCH, or how have they been able to take advantage of the chipset where others have not? It seems to me the canterwood was built with indigenous memory enhancemenst called PAT, yet so few are actually able to take advantage of PAT, except 2nd-teir memory makers? I'm not technically inclined to trace down the source, of the anomalies, if there is any, but it's become obvious to nme having the IC7-G, IS7-E, and 4PCA3+ PAT features on all three have yet to produce anytjing but errors for me. it's like "Bizzaro World" up is down, and down is up.

[Liquid3D]

Originally posted by Zroc
Purty scores, Liquid ;)
By the way, you can't disable PAT on an 875P chipset (what the BIOS is refering to is not actual PAT, but some other functions, similar to Game Acceleration on Abit, etc). PAT is on with an 875P regardless...the only way to shut it off is sticking and 865PE in there... Zroc, why then does Sandra indicate it's disabled here when I Disabled "Intel fast CS"? Yet in the next post down when I enabled "Intel fast CS" Sandra concurs?

[Liquid3D]

And here where I enabled the "Intel fast CS" it indicates it's activated?

[Zroc]

Because Sandra (and CPU-Z) reports it incorrectly. They'll tell you there's no PAT on Intel's own 875P as well, but it's on.

[FragMagnet]

Originally posted by Kalway
By the way theoretical bandwidth on Liquid3d's setup is ~8700mb/sec. He's getting ~6150mb/sec (rounding up for easier math). That's 70.6% efficiency. At 2-2-2-5 you get ~95% efficiency. My point is there is actually less of a bottleneck when using the divider since the memory is sending data faster. Run the memory at 200mhz with 2-2-2-5 DCDDR and see what your bandwidth is. It should be ~6000-6200mb/sec (Intel's chipsets are annoying with their efficiency).

IF my calculations are correct at 240 2-2-2-5 you SHOULD get easily over 7000mb/sec. If one of you with the memory + board would be kind enough to try this please post your results. Try 1:1 and 5:4 if you can also try with PAT enabled. I'd get some myself but there are no test boards available right now :( so I can't do it. We also don't have any memory that can reach that speed here haha.

Post with some results.
Nice fsb Liquid

I did, using XtremeDDR @ 1:1 @ 249 3-4-3-3 was worth 6130 in Sandra (IC7....025 bios) Now using the same cpu at the same overall speed, and dropping the divider down to 5:4 and 3:2, the Sandra scores went through the floor, but using a non overclocked 9800P, all 3 offered 3dM2K1 scores within 200 points.
I didn't have both types of DDR at the same time but had someone in another forum compare 2-6-2-2 to my 2.5-7-3-3 @ 260 5:4 and there was exacty 30-50 Sandra mem points difference.
BUT !!
With this Corsair, @ 2-2-2-5 @ 5:4 I was now able to turn on all the options below Game Accelerator, turn on the FW's in SmartGart and run the cpu @ 258 fsb 5:4, that was worth an extra 1000 points in 3dM2K1 :D
So giving up the Sandra #s were worth 3dm2k1 score in the long run.
I could run 257 fsb 5:4 with the XtremeDDR but with the 2.5-7-3-3, and noe of the mem options enabled or FW

Just my experience. you'll see a lot of reviews trying to compare these things with ES samples, jacking the multi all over the place, comparing a cpu with the higher fsb and different mem timings to a cpu with slower.......etc, etc. I couldn't find one review, regardless of the brand of mem they were testing, to be an apples to apples comparison. So I bought the stuff myself and in the end was happy ;)

[eva2000]

Originally posted by Liquid3D
[B
I guess the question is, what is A-DATA using, or doing which allows their memory to either circumvent the problems innate to the Canterwood's MCH, or how have they been able to take advantage of the chipset where others have not? [/B] maybe they're doing the same thing that OCZ does with their process ?

[DDTUNG]

I don't understand all the fuzz about this ram. I have some ADATA DDR500 ram that runs at 280 DC 1:1 at 2.85V, 3-4-4-8 timings, with a 2.4C on air and a P4C800 with PAT on. I recall Sandra memory benches in the 6500s. And it is stable enough folding gromacs 24/7. Cost was under $80 per 256MB stick.

DDTUNG:cool:

[Zroc]

Originally posted by DDTUNG
I don't understand all the fuzz about this ram. I have some ADATA DDR500 ram that runs at 280 DC 1:1 at 2.85V, 3-4-4-8 timings, with a 2.4C on air and a P4C800 with PAT on. I recall Sandra memory benches in the 6500s. And it is stable enough folding gromacs 24/7. Cost was under $80 per 256MB stick.

DDTUNG:cool:
DDTUNG, will that ADATA actually do good timings at lower speeds? Like, will it do 2/2/2/5 at DDR400 even? If it WILL do good timings, what's the highest you've done with 'em?

[eva2000]

Originally posted by Zroc
DDTUNG, will that ADATA actually do good timings at lower speeds? Like, will it do 2/2/2/5 at DDR400 even? If it WILL do good timings, what's the highest you've done with 'em? depends on the board and it's bios too i.e.

i can do 2 x 256MB ADATA PC4000 @233mhz 2.5-4-3-7 at 2.85v on PX865PE PRO2 but so far can't do the same on 4PCA3+ same vdimm or higher

[Zroc]

Originally posted by eva2000
depends on the board and it's bios too i.e.

i can do 2 x 256MB ADATA PC4000 @233mhz 2.5-4-3-7 at 2.85v on PX865PE PRO2 but so far can't do the same on 4PCA3+ same vdimm or higher

So it's like all this other PC4000 stuff...won't do good timings, period.

[Shane5578]

i really doubt that we're going to get anymore new ddr memory that can do really tight timings.

i hear that ddr2 will be like cas 5

[daddy_fizz]

Originally posted by Zroc
DDTUNG, will that ADATA actually do good timings at lower speeds? Like, will it do 2/2/2/5 at DDR400 even? If it WILL do good timings, what's the highest you've done with 'em?

a review i read somewhere about a-data said that the best they could do was 2-3-3-5 at DDR400...

~Fizz

[STEvil]

This makes me want to sell my BH-6 sticks for $$$ and just buy some cheap ram to hold me over to DDR2... hehe 8-)

[eva2000]

well here's my 2.4C Epox 4PCA3+ with ADATA with Intel Fast = Enabled for dual channel and Disabled for single channel and CPC = AUTO

1x 256MB ADATA PC4000 @ 282mhz 3-4-4-8 at 1:1 at 3.2v

http://www.fileshosts.com/pentium4/D1/epox/benchmarks/2.4C_3/adata/pc4000/1x256/282-282-3448-ifcpc-off/cpuz.jpg

2x 256MB ADATA PC4000 @ 275mhz 2.5-4-4-7 at 1:1 at 3.2v

http://www.fileshosts.com/pentium4/D1/epox/benchmarks/2.4C_3/adata/pc4000/2x256/275-275-2.5447-ifcpc-off/cpuz1.jpg

http://www.fileshosts.com/pentium4/D1/epox/benchmarks/2.4C_3/adata/pc4000/2x256/275-275-2.5447-ifcpc-off/mem-buff-ifon-cpcauto.jpg

(strange clevor with ADATA PC4000 and 4PCA3+ got much higher unbuffered bandwidth 300-700MB+ more)
http://www.fileshosts.com/pentium4/D1/epox/benchmarks/2.4C_3/adata/pc4000/2x256/275-275-2.5447-ifcpc-off/mem-unbuff-ifon-cpcauto.jpg

http://www.fileshosts.com/pentium4/D1/epox/benchmarks/2.4C_3/adata/pc4000/2x256/275-275-2.5447-ifcpc-off/pcmark2002-ifon-cpcauto-490-330.jpg

http://www.fileshosts.com/pentium4/D1/epox/benchmarks/2.4C_3/adata/pc4000/2x256/275-275-2.5447-ifcpc-off/aida-read-ifon-cpcauto.jpg

http://www.fileshosts.com/pentium4/D1/epox/benchmarks/2.4C_3/adata/pc4000/2x256/275-275-2.5447-ifcpc-off/aida-write-ifon-cpcauto.jpg

http://www.fileshosts.com/pentium4/D1/epox/benchmarks/2.4C_3/adata/pc4000/2x256/275-275-2.5447-ifcpc-off/sciencemark-ifon-cpcauto.jpg

[Liquid3D]

Dam nice numbers from that memory! Definately loves VDIMM though doesn't it? Hey what benchmark program is that at the very bottom? Is that CPU-Z? Does it run benchamrks?

[eva2000]

last screen is the sciencemark 2.0 app i keep on talking about ;)

i uploaded it to http://fileshosts.com/files/sciencemark/ScienceMark2.zip for you in another thread here when i was explaining sciencemark :D

[STEvil]

Someone should try pumping 3.6v into that A-Data like the old BH-6 stuff and see if they can hit 300mhz....

[eva2000]

they'd need a cpu that could do 300FSB heh.. me needs on of those :D

[Creative]

Originally posted by eva2000
they'd need a cpu that could do 300FSB heh.. me needs on of those :D
I have 1 ;)

How does the 3dmark compare with higher fsb eva? Cos atm Im still not impressed with this fsb while the timings soft....like for example, I beat your sandra unbuffered by 700 MB/s with 245fsb tight timings.....so Im wondering what the comparion would be for 3dmark etc?

[Shane5578]

hey eva2000 which bios are you using?

[eva2000]

latest official epox bios July 3, 2003 ftp://ftp.epox.com/motherboard/support/BIOS/4PC33703.exe

[Liquid3D]

PM me Eva I have an interesting proposal for ya. If your close to me I have a 2.4C that does 300FSB easy, maybe i could ship it, let you borrow it? Thanx, I finally got the SciMark2 to run. I like all the info it gives. After-all, for those of us concerned about Bandwidth, shouldn't we be concerned about Cache perfromance as well? This the OCZ GolD400 @ 251, but everything was running when I did it, including USB Internet, and bunch of other programs.

[QuadDamage]

well, it's a TwinMOS at 262fsb 1:1 at 2.5/3/3/8 and CPC enabled:

http://www.theforumisdown.com/uploadfiles/0103/0_262_11.JPG

edit : also on Epos 4pca3+.

[Liquid3D]

Nevermind the message above Eva, I see your an Aussie, too far to ship you anything.

Here's an interesting question. Since people are getting these great bandwidth's with Kinston Hyper-X3500 (the only 7K+ MB/s I'd seen) I thought I try some older Corsair which must use simliliar IC's. So I have in some XMS3500. I cranked up the VDIMM to 3.0 (and even 3.@V at 280FSB) and the highest I got was 250FSB at 2-6-2-2, and the benchies were horrific. So I dropped the 240FSB and tightened the timimgs to 2-5-2-2, and even enabled Intel Fast CS, but look at these horrific benchmarks. I keep getting 71% efficiency no matter what? Could there be some other bottleneck I'm not aware of? (Oh maybe it's DIMM's 1&3 but still);

[eva2000]

yeah in Oz bit too far, but thanks for the thought :)

my 2x 512mb xms3500c2 perform the same on 4PCA3+ i think the epox bios memory performance isn't as tweaked as IC7 or PX865PE PRO2, the performance gain from enabling Intel Fast is less than 0.5-1.0%

see here http://www.xtremesystems.org/forums/showthread.php?s=&postid=198227#post198227

[STEvil]

Oz isnt too far...

I shipped my old 8500LE that did 323/325 stock over there for $20 by air (dont ship by boat!).

[Liquid3D]

Really, only $20 smackers eh? Eva you said above "I need a CPU that'll do 300FSB..." I have two, but were you serious? Your 2.6C or 2.4C will do 300FSB? You have several mobo's, although I haven't been able to reach 300FSB on my Epox, but my IS7-E did in dual channel everytime. I don't think I reached 300FSB on my IC7-G either? Anyway, i was thinkling of maybe letting you use a 2.4C which has done 300FSB on my Abit IS7-E so you could test that 300FSB theory with your A-DATA. Let me ponder it.

[jathlon77]

Where is PAT in the bios for Asus P4c800 delux? I cant find it. What does it stand for also?
Thx

Originally posted by Shane5578
fast cs should be enabled and cpc disabled.

otherwise cpuz/ctiaw will show PAT disabled

cpc only works near stock speeds for me

[lalPOOO]

Whats the point of brining up a 5+ month old thread? Just make a new one. (actually I think you already did...)

[PyroTeknik]

"strong bad back from the dead o holy crap......"

-homestarrunner

[Shane5578]

Originally posted by jathlon77
Where is PAT in the bios for Asus P4c800 delux? I cant find it. What does it stand for also?
Thx

i was referring to the epox 4pca3+ where you can enable pat.

p4c800 has pat enabled all the time. you can't change in the bios

[Liquid3D]

What a circle of life we gone through. Here we are months later and the only thig that has changed are the names. My first Intel board for the 2.4C was the Abit IS7-E, and I've gone through the Epox 4PCA3+, Abit IC7-G, Asus P4C800E-Deluxe, and now closing the circle with the AI7. How did the name change? Well I got so bored with the Asus I wanted to try something new. SO I traded a $350 kit of OCZ EL4200 for this board. The best thing that came out of this was that I got to trade with a firend, and I hope he enjoys the memory. As for myself, this board isn't too bad. It offers 3.20VDIMM, although the bandwidth between this and the Asus P4C800Ed are about 350MB/s diff. The board has some interesting features. I didn't really think I'd need Guru, and I was criticizing it, then I discovered it's not just a watered down version of ClockGen. It has VDIMM, VDDQ, And Vcore adjustability from within windows along with it's FSB slider.

I wonder if it defeats the purpose if it's made for the beginning Overclocker, as one has to learn these BIOS adjustmensts anyway? Memory Ratio, GAT settings, these are not accessable from the Windows guru utility, although it does feature some very decent monitoring features. here's a screenie with OCZ PC4400 at 275FSB, I've been unable to get past memory errors even at 3.10V DDR 285FSB was re-booting from SiS Sandra Memory bench;

[nailbomb]

Where the hell does one BUY ADATA sticks in the USA? I have allways heard good things about them, but never seen any ;)

[st0nedpenguin]

Originally posted by Shane5578
i was referring to the epox 4pca3+ where you can enable pat.

p4c800 has pat enabled all the time. you can't change in the bios

You can't?

Then how come disabling it in the BIOS makes CPU-Z read it as disabled?

[Liquid3D]

I definately believe certain aspects of PAT can be disabled. In the Asus P4C800Ed BIOS (and most others) there's two places where PAT can be effected. First in the Jumperless, or SOftmenu basically the primary menau where adjustments are made to FSB, Boot strap, AGp, Vcore, and DDR voltage. At the bootm ofthe menu it usully gives the option, PAT = Standard, Turbo, Auto.

The second location is in the Chipset Features or mmory elated menu. There it simply says PAT - Enabled or Auto. So while it doen't necessarily state Disabled, placing PAT on Auto in both these menu's then has SiS Sandra reading it as follows; "Performance Acceleration Tech. No" As seen below;

[st0nedpenguin]

Yeah, that's what I was thinking, if I leave the settings on auto, PAT seems to be disabled when overclocking, but enabled at stock.